首席AI官的角色解构：科学家、架构师与教练的三维演进

Hello everyone. How are you? OK. I'm so glad to see you all here, still upbeat and ready. So, today I wanted to share my experience as an AI director. This is a very new role. I get asked a lot of questions about her. I have been in this position for almost 2 years, and in various variations for the last five years. And I would like to start by understanding who is gathered in the hall today. How many people here are from companies that develop and sell software? Perfectly. And how many of you have an AI leader who isn't a CEO or CTO ? Very interesting. So we'll get straight to the point. This helps.

As I said, this role is new, but it is becoming increasingly popular. And I found out that it means completely different things depending on two criteria. First, it's the type of company you work for. Secondly, it is the company's level of maturity in the field of AI. There is another dimension that really depends on the skills and experience of the person in the position. I've seen that it also varies a lot. In some cases, a position was created for a specific person with a specific skill set, right? In other cases, a very specific need arose, and a person was hired to meet it. Since it is difficult to find a specialist for this position, we see that companies often show flexibility in responsibilities, adapting them to the candidate to make the most of their strengths. So I've seen how sometimes this role is divided into two . The most interesting thing is that even IBM studies conducted a year apart show growth: 11% in 2024, 26% in 2025, and now 76%. Em. Okay .

The difficulty with this role is that it sounds great. Okay , I have a director of artificial intelligence, but AI is everywhere, right ? From how you use AI to be more productive, to compliance and safety, selecting and evaluating vendors, creating customer personas, product development. It affects everything. So this title doesn't mean anything until you specify what kind of company it is and where it is in its journey. The way I like to think about this role, and how I've seen it develop, is as a combination of three different roles. It's a scientist, it's an architect, and it's a coach.

A scientist is when you research, experiment, and build. An architect also builds, but he is much more involved in strategy. If the company sells software products, you think about creating new software and influence the product strategy. If the company doesn't create software, you implement AI in a way that doesn't disrupt what's already being done, right? It all depends on where you are on your AI journey. If you're at the very beginning, you think about how to use AI to change that. I come from a product company, so my approach has been from the previous one. So part of my job is to be responsible for our AI initiatives and how they help the company and customers.

Finally, the role of a coach came as a surprise to me. I worked for 20 years at IBM Research, leading a third of the global AI research organization. When I first took on this role, I thought everyone in a software company was familiar with AI and loved it. It was a complete surprise to me to find that this is not always the case. Some people are afraid of this. Some people don't know where to start. So this coaching role was unexpected, but very interesting. And so you'll see a lot of people in those positions with a wide range, and I think it depends on the type of company and the skills you bring to the role. So, for a scientist, even at One Medical, you don't necessarily have to have a PhD, you just do intensive research. I believe that in the world of AI, which is evolving so rapidly, you should always have a part that is 3, 5, 10 years ahead, while others are working on what is happening right now, otherwise it will just fly past you, right?

And again, you can also have people on that journey who are more on the architect side. You have a lot of people like that. In this case, there are also companies that have an AI program manager, so they have an AI project management office. You will see that they don't even have to have technical expertise. All they do is they manage the projects and they report, right? So, this doesn't require deep knowledge of AI, but it does require a good understanding of the business and being a very effective operations manager, right? On the other hand, what I do: you shape the entire strategy of the company, rethink how we work, in terms of productivity, as well as rethink our products and develop new ones.

So, one person who has been in this role for quite some time, his name is Richard Freeman, did an interesting study, surveying many people in this position. He was also a consultant for companies looking for someone to fill this role. He talked to recruiters, asked questions, analyzed what the market thought about it. And here are some of the areas he discovered . This confirms that it really all depends on the person.

The rows are the personality type and skill set, and the columns are the “sliders,” right? I will give examples from my own experience, as this field is so changing and new, and I hope this will be useful to you too. As you can see, this doesn't mean that one person should do everything. It depends on you. What are you best at? What do you want to learn? What do you want to focus on? And what does the company need? And if there is something you don't want to do, or something you are not good at, it is very important to see how you can work with other people who will close this gap, who are strong in these things, right?

So, going back to the sliders and all that , I would say that at WSO2, where I work now, I'm about 20% scientist, 60% architect, and 20% coach. Um, and these sliders below also show what level I'm at in each dimension, right? So, we have a small research group there with whom we have done some work on building, on agents, on how to write code, and how to verify it. But mostly, as I said, I'm an architect, which means looking at the big picture: how we design things, what we build. And then coach, I'll tell you about it later.

About 75% of the people in the company are technical specialists, and they are extremely curious, you know? This is the kind of place where if you're not curious, you probably won't stay long, right? Um, so I don't have to do much internal coaching. I do a lot of external coaching. Um, but in my previous job, I had to do a lot of internal coaching, which was also interesting.

So, when we look at internal coaching, you need to understand that developers use AI differently. For example, are they using code generation? We've already had this conversation today, you know, measuring acceptance rates is a tricky thing. People can accept suggestions and then immediately reject them. So you shouldn't just rely on that. I've heard some say that we need 80% adoption. Why? Why do you want people to accept code they don't need? And what if your developers are writing code that's already in the AI's memory and shouldn't even be writing it? What are you measuring? You're measuring adoption rates, but you don't know what you're actually measuring. So someone asked me today, “What do you measure instead?” "And I said, 'Okay, that's what I'm looking at.'"

Yes, on the level of AI proficiency among employees. Is it just a few people who know AI in the competency center? And whenever someone needs something, you should reach out to these people. Are there developers in different departments? Are there people who know enough to do their jobs and actually use it, or is it widespread? And are there full-stack AI developers who can use this, create it, and change it? So coaching was about getting people to a level where they could be creative, so they weren't afraid of it, and they were familiar enough with it that they could start being creative.

The other part of coaching is internal, is eating your own dog food. I heard an interesting thing about it today. Someone said to me, "Oh, I'm at this company, I wo n't name it , and they're forcing us to use their product. And it would be great if no one ignored feedback, right? To improve the product." So hopefully we'll close the feedback loop and not just force people to use ours, but they'll actually use it to make it better. And the other part is external coaching, as I said, talking to customers, talking at conferences, explaining to people how they should think about using AI.

So, as an architect, what do you think about? In a product company, this is product strategy. Last year, I focused on adding AI features to all of our products. In addition, you need to understand that the product is not just the software itself. It's the whole ecosystem. It's the user experience, but it's also the documentation. Now you should have documentation for people, but people aren't the only ones reading your documentation. They are read by LLMs. So the documentation should be suitable for LLM. Last year, we added world-class agent support, LLM, and tools to all of our products. And this year, we are making sure that every product can be used by agents. So, everything must have an MCP- server, skills, CLI, etc. Um, agent-proof pricing. What is this? This is what you should think about as an architect.

And as for internal productivity, and here's what to look at: What are the bottlenecks? Where are you wasting time? That's what you should measure. Once you start measuring something, everyone will try to optimize it.

So, a few words about WSO2. We are about 20 years old. Our annual revenue (ARR) is around 150 million. Our software powers many systems in over 90 countries on six continents. Our core products and our history are API platforms, integration, and identity management. Over the last two years, we have made sure that we have AI in our API gateways, identity, and so on. We have made sure that people who use APIs can manage APIs for AI, but can also use AI to manage and create APIs. The same goes for integration.

We also looked at identity management. When you think about agents, how do they log into systems? How do they get access to things? How do you monitor who accesses what? This is what we have in identity management too. We added agent identification. This was all last year. And this year we released a platform for agents. um, it allows you to use these things and also bring it all together and manage the full lifecycle of these agents.

So, I will give a few examples from scientists as well. We have a small research group, as I said. A lot of that work is in evaluation, but also in code. How do you verify code? How do you write tests? How do you make sure the models actually do what they are supposed to do? And doing things like this helps close that loop and make sure that we have strong research behind the products that we're putting out.

So, I think I'll wrap it up here. What I wanted to convey to you is that there is no one-size-fits-all solution. It all depends on the company. You have to work very closely with others. My advice if you try something like this is to have a very strong support and relationship with the CEO because you will have to work very closely with many other departments in the company. Also, look at what you are good at, what you like to do, what you can get paid to do, and try to find something at the intersection of that. Since this role is still very fluid, try to understand whether it interests you and suits you, try to shape it so that you are at the center, so that it brings you joy, and what you don't like, try to transfer to someone else or maybe another company will be a better fit, where your interests, preferences, needs and strengths will better match the real tasks . Okay, thank you very much. If you have any questions, I'm here.